AI Model Card
A documentation template describing the details, limitations, and intended use of an AI model.
Active Learning
A process where the model selects the most informative data points to be labeled by a human.
Algorithm
A step-by-step procedure for solving a problem or performing a computation.
Anomaly Detection
Identifying unusual data points that differ from the norm.
Artificial General Intelligence (AGI)
A type of AI with the ability to understand, learn, and apply knowledge across a wide range of tasks.
Artificial Intelligence (AI)
The simulation of human intelligence processes by machines, especially computer systems.
Artificial Narrow Intelligence (ANI)
AI that is specialized in one specific task.
Artificial Superintelligence (ASI)
A hypothetical AI that surpasses human intelligence across all fields.
Attention Mechanism
Technique allowing models to focus on relevant parts of input data, essential in transformer models.
AutoML (Automated Machine Learning)
Tools that automate the end-to-end process of applying ML to real-world problems.
BERT (Bidirectional Encoder Representations from Transformers)
A transformer-based model that understands context in both directions.
Backpropagation
A method used to train neural networks by updating weights based on error.
Bias
Systematic error introduced by an assumption in the machine learning process.
Bias Mitigation
Techniques used to reduce bias in AI models.
Black Box
A model whose internal workings are not visible or understandable.
Chatbot
A software application used to conduct an online chat conversation via text or text-to-speech.
Classification
Assigning inputs into predefined categories.
Clustering
Grouping similar data points together in unsupervised learning.
Computer Vision
A field of AI that trains computers to interpret and understand the visual world.
Concept Drift
When the statistical properties of target variables change over time, affecting model performance.
Data Augmentation
Techniques used to increase the amount and diversity of data.
Data Preprocessing
Steps taken to clean and prepare data before training a model.
Deep Learning
A type of machine learning using neural networks with many layers.
Diffusion Model
A generative model that learns to create data (like images) by reversing a noise process.
Dimensionality Reduction
Techniques to reduce the number of input variables in a dataset.
Embedding
A representation of text or data in a dense vector space.
Embeddings Store
A searchable database of vector embeddings, used in semantic search and retrieval-augmented generation.
Epoch
One complete pass through the entire training dataset.
Ethical AI
AI developed and deployed in a way that respects human rights, fairness, and accountability.
Explainable AI (XAI)
AI systems designed to explain their decisions to humans.
Feature Engineering
The process of selecting and transforming variables to improve model performance.
Federated Learning
Training machine learning models across decentralized devices or servers.
Few-shot Learning
Learning from a small number of examples.
Fine-tuning
Training a pre-trained model on a specific task or dataset.
GPT (Generative Pre-trained Transformer)
A generative language model developed by OpenAI.
Generative AI
AI systems that can create new content, such as text, images, or music.
Gradient Descent
An optimization algorithm used to minimize the loss function in training.
Guardrails (in AI)
Mechanisms to enforce ethical, safety, or behavioral constraints in AI models.
Hallucination (in AI)
When an AI generates output that is plausible but factually incorrect or nonsensical.
Hyperparameter
Settings used to control the training process of a model.
Inference
The process of using a trained model to make predictions.
Knowledge Graph
A structured representation of facts, entities, and relationships used for reasoning and inference.
Large Language Model (LLM)
A type of generative AI trained on vast amounts of text to understand and generate human-like text.
Latency (in inference)
The time delay between inputting data and receiving a model’s output.
LoRA (Low-Rank Adaptation)
A technique for efficiently fine-tuning large language models using fewer resources.
Loss Function
A function that measures the error between predicted and actual outcomes.
MOE (Mixture of Experts)
A neural network architecture that selectively activates parts of the model for efficiency and performance.
Machine Learning (ML)
A subset of AI that involves the use of algorithms and statistical models to enable machines to improve at tasks with experience.
Model
A mathematical representation of a real-world process, trained to make predictions or decisions.
Model Distillation
The process of creating a smaller model that mimics a larger, more complex one.
Multimodal AI
AI systems that understand and process multiple data types (e.g., text + image).
Natural Language Processing (NLP)
A field of AI that gives machines the ability to read, understand, and respond in human languages.
Neural Network
A network of artificial neurons that mimic the human brain’s structure to process data.
Overfitting
A modeling error which occurs when a model is too complex and captures noise in the data.
Prompt Engineering
Crafting input prompts to effectively interact with language models.
Prompt Injection
A vulnerability where malicious prompts are embedded in input data to manipulate AI behavior.
RLHF (Reinforcement Learning from Human Feedback)
A training method that aligns model outputs with human preferences.
Regression
Predicting continuous values from input data.
Reinforcement Learning
A type of machine learning where agents learn to make decisions by receiving rewards or penalties.
Retrieval-Augmented Generation (RAG)
Combines language models with a retrieval system for better-informed responses.
Self-supervised Learning
A method where the model generates its own labels from raw data during training.
Supervised Learning
A type of machine learning where models are trained on labeled data.
Synthetic Data
Artificially generated data used to train AI models.
Token Limit
The maximum number of input and output tokens a model can process at once.
Tokenization
The process of splitting text into smaller pieces (tokens), often words or subwords.
Training Data
The dataset used to train an AI model.
Transfer Learning
Using a pre-trained model on a new, but related problem.
Transformer
A deep learning architecture that uses self-attention; forms the backbone of models like GPT and BERT.
Turing Test
A test to determine whether a machine can exhibit human-like intelligence.
Underfitting
A scenario where a model is too simple to capture the underlying pattern of the data.
Unsupervised Learning
Machine learning using data that has not been labeled or categorized.
Variance
The model’s sensitivity to small changes in the training dataset.
Vector Database
A specialized database optimized for storing and searching high-dimensional vectors.
White Box
A model whose decisions and internal logic are transparent.
Zero-shot Learning
Making predictions without the model having seen any examples of the task.